counterparty-channel-discipline

Pass

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is specifically designed to manage agents that ingest and process data from untrusted external environments such as shared workspace channels and direct messages.
  • Ingestion points: As described in SKILL.md, the agent is exposed to messages from "external counterparties," "shared channels," "unknown DMs," and "group chats."
  • Boundary markers: The skill provides a robust mitigation strategy via the references/strict-prompt.template.md, which defines an immutable set of instructions forcing the agent to verify participation consent and ignore historical thread context when determining if it should respond.
  • Capability inventory: The instructions reference the agent's ability to perform network-based messaging actions (send, edit, stream) via a runtime adapter, which the skill attempts to govern through strict output classification rules.
  • Sanitization: The skill explicitly clarifies that "Output classification is not text sanitization," indicating a reliance on higher-level transport guards for final content safety while focusing its own rules on logic and context containment.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 10, 2026, 01:58 PM
Security Audit — agent-trust-hub — counterparty-channel-discipline